AI agents are bad at long-term choices, Microsoft-backed test finds
Researchers from Microsoft and City University of Hong Kong built a test of how well AI agents make long-term choices. They call this skill "taste." Each question asks a model to pick the better of two paths in a task. The best model, GPT-5.6 Sol, got 59.7 percent right. Pure guessing would score 25 percent.
Where this came from. One outlet reported this. We read it, then wrote the sixty words above.